I have a phenotypic trait I am interested in, and a number of SNPs have been reported in literature to be associated with the trait (GWAS studies).
Each SNP is associated with the trait with its own estimated effect and p-value. I would like to combine them in an overall index, so that based on the genotypes of a new subject at those SNP loci, I could come up with a genetic value/score for that individual for the trait.
The first step could simply be to add the estimated SNP effects, but in this way I would not take into account the different significance with which the SNPs are associated with the trait. Alternatively, I could weigh the SNP effects by their p-value (or inverse of).
Any other ideas? Examples from literature?
Thanks!
You may like to have a look at doi:10.1038/nature08185 where the authors developed a method for creating a multi-marker risk score
(stratified by p-value of single SNPs and weighted with effect size)
Wouldn't be easier and more straightforward to define the haplotipes? P-values are sample dependent, therefore, if your cohort is not structured the exact same as the discovery cohort, the P-values not necessarily be consistent, it gets worse as the effect size and allele frequency gets smaller.
I think you should enrich your GWAS data with other biological information like Chromatin Accessibility from ENCODE and others. There are many web tools to do this: RegulomeDB and GWAS3D are the ones that I can remember at the moment.
You could also try to find if the SNPs are within a TFBS by using the JASPARdb or MATCH from TRANSFAC or direcly from ChIP-seq data from ENCODE (though its limited). Once you have enriched your data, I would run machine learning algorithms or PCA to create the model.
So as Isain, I believe whithout having the knowledge about interallelic interactions of the test SNPs ( i.e. LD map of the locus and haplotypes) such meta-analysis wouldn't be really profitable. Meanwhile, this post from the "Getting Genetics done" might be helpful.
http://gettinggeneticsdone.blogspot.com.tr/2010/01/gwas-manhattan-plots-and-qq-plots-using.html?m=1
Hi Filippo,
Note that effects from GWAS are most likely overestimated, because LD with other SNPs is not taken into account, and I don't think p-values is a good manner to weight SNPs as they are sample dependent and do not properly account for false discovery rate.
I don't think you will get a large enough accuracy if you do not estimate the effect of the SNPs in your population with a proper multimarker model such as Bayes A, B, C, R, Lasso or GBLUP, and using a large reference population.
- Badges
- Science topic
- Similar topics
- Bioinformatics
- Bioinformatics and Computational Biology